TensorFlow 2.0 Beta is available Learn more

Release notes

Nightly

New datasets

Features

  • Add in_memory option to cache small dataset in RAM.
  • Better sharding, shuffling and sub-split
  • It is now possible to add arbitrary metadata to tfds.core.DatasetInfo which will be stored/restored with the dataset. See tfds.core.Metadata.
  • Better proxy support, possibility to add certificate
  • Add decoders kwargs to override the default feature decoding (guide).